Hi devs,
and advanced users. Has anyone perforned, or has plans to implement, non-linear (multi-) regression in GRASS (not through R, of course)?
Thank you, Nikos
Hi Nikos,
On Sat, Feb 21, 2015 at 2:08 PM, Nikos Alexandris
<nik@nikosalexandris.net> wrote:
and advanced users. Has anyone perforned, or has plans to implement,
non-linear (multi-) regression in GRASS (not through R, of course)?
I did a pixel by pixel regression using most of the code of the
v.class.ml module, but the approach still need to act manually.
So it is not ready...
During next spring I would like to make a bit of order on the
v.class.ml module, maybe change also the name, to allow supervised,
unsupervised and regression tasks through scikit-learn library. If you
have some ideas on how would you like to organized these
module/modules, like: which inputs are required, the outputs, the
optional options and steps, please share them and we can start to
discuss on how implement what.
Best regards.
Pietro
On Sat, Feb 21, 2015 at 3:21 PM, Pietro <peter.zamb@gmail.com> wrote:
Hi Nikos,
On Sat, Feb 21, 2015 at 2:08 PM, Nikos Alexandris
<nik@nikosalexandris.net> wrote:
> and advanced users. Has anyone perforned, or has plans to implement,
> non-linear (multi-) regression in GRASS (not through R, of course)?I did a pixel by pixel regression using most of the code of the
v.class.ml module, but the approach still need to act manually.
So it is not ready...During next spring I would like to make a bit of order on the
v.class.ml module, maybe change also the name, to allow supervised,
unsupervised and regression tasks through scikit-learn library.
If you
have some ideas on how would you like to organized these
module/modules, like: which inputs are required, the outputs, the
optional options and steps, please share them and we can start to
discuss on how implement what.
The v.class.ml module is for vector classification. Perhaps you already had
this is mind, but would it be possible to implement something similar for
rasters, similar to the r.regression.multi module (but possibly allowing
the use of point vector data as calibration data)?
Best regards.
Pietro
_______________________________________________
grass-dev mailing list
grass-dev@lists.osgeo.org
http://lists.osgeo.org/mailman/listinfo/grass-dev
On Sat, Feb 21, 2015 at 2:08 PM, Nikos Alexandris
<nik@nikosalexandris.net> wrote:
Hi devs,
and advanced users. Has anyone perforned, or has plans to implement,
non-linear (multi-) regression in GRASS (not through R, of course)?
If it is raster based, you can use r.regression.multi to perform
non-linear (multi-) regression in GRASS. It depends on what you mean
with non-linear. E.g. if you want to perform a second-order multiple
regression with 2 predictors, the formula is in R syntax
y ~ x1 + x2 + x1 * x2 + x1 * x1 + x2 * x2
You can calculate the terms x1 * x2, x1 * x1, x2 * x2 with r.mapcalc
and feed everything to r.regression.multi.
logarithmic:
y ~ log(x1)
r.mapcalc "logx1 = log(x1)"
exponential:
y ~ exp(x1)
r.mapcalc "expx1 = exp(x1)"
A combination of say a 4th-order polynomial with exponential
components and 10 predictors could be constructed accordingly.
HTH,
Markus M
Thank you, Nikos
_______________________________________________
grass-dev mailing list
grass-dev@lists.osgeo.org
http://lists.osgeo.org/mailman/listinfo/grass-dev
Nikos Alexandris:
and advanced users. Has anyone performed, or has plans to implement,
non-linear (multi-) regression in GRASS (not through R, of course)?
Pietro:
I did a pixel by pixel regression using most of the code of the
v.class.ml module, but the approach still need to act manually.
So it is not ready...
During next spring I would like to make a bit of order on the
v.class.ml module, maybe change also the name, to allow supervised,
unsupervised and regression tasks through scikit-learn library. If you
have some ideas on how would you like to organized these
module/modules, like: which inputs are required, the outputs, the
optional options and steps, please share them and we can start to
discuss on how implement what.
Pietro,
thanks for chiming-in. I need an approach for raster images, as (usual, perfectly) explained by Markus M. I think, it's exactly what I am in need for.
Nikos
Paulo van Breugel wrote:
..
The v.class.ml module is for vector classification. Perhaps you already had
this is mind, but would it be possible to implement something similar for
rasters, similar to the r.regression.multi module (but possibly allowing
the use of point vector data as calibration data)?
Paulo,
what's exactly the idea of using vector data for calibration? Can you write one more sentence about it?
Thanks Nikos
Nikos Alexandris:
Hi devs,
and advanced users. Has anyone performed, or has plans to implement,
non-linear (multi-) regression in GRASS (not through R, of course)?
Markus Metz:
If it is raster based, you can use r.regression.multi to perform
non-linear (multi-) regression in GRASS. It depends on what you mean
with non-linear. E.g. if you want to perform a second-order multiple
regression with 2 predictors, the formula is in R syntax
y ~ x1 + x2 + x1 * x2 + x1 * x1 + x2 * x2
You can calculate the terms x1 * x2, x1 * x1, x2 * x2 with r.mapcalc
and feed everything to r.regression.multi.
logarithmic:
y ~ log(x1)
r.mapcalc "logx1 = log(x1)"
exponential:
y ~ exp(x1)
r.mapcalc "expx1 = exp(x1)"
A combination of say a 4th-order polynomial with exponential
components and 10 predictors could be constructed accordingly.
Speechless. Nikos
On Tue, Feb 24, 2015 at 3:31 PM, Nikos Alexandris <nik@nikosalexandris.net>
wrote:
Nikos Alexandris:
and advanced users. Has anyone performed, or has plans to implement,
non-linear (multi-) regression in GRASS (not through R, of course)?
Pietro:
I did a pixel by pixel regression using most of the code of the
v.class.ml module, but the approach still need to act manually.
So it is not ready...During next spring I would like to make a bit of order on the
v.class.ml module, maybe change also the name, to allow supervised,
unsupervised and regression tasks through scikit-learn library. If you
have some ideas on how would you like to organized these
module/modules, like: which inputs are required, the outputs, the
optional options and steps, please share them and we can start to
discuss on how implement what.Pietro,
thanks for chiming-in. I need an approach for raster images, as (usual,
perfectly) explained by Markus M. I think, it's exactly what I am in need
for.
I wonder if it is possible to utilize the scikit-learn library for rasters
as well.. The library offers a broad range of regression models, as well as
clustering and classification functions.
Nikos
_______________________________________________
grass-dev mailing list
grass-dev@lists.osgeo.org
http://lists.osgeo.org/mailman/listinfo/grass-dev
On Tue, Feb 24, 2015 at 5:30 PM, Paulo van Breugel
<p.vanbreugel@gmail.com> wrote:
I wonder if it is possible to utilize the scikit-learn library for rasters
as well.. The library offers a broad range of regression models, as well as
clustering and classification functions.
yes, I've done it.
I've applied severl regression models for more than 300 rasters (as
features) and it was fast but the region was quite small.
Pietro